Professional Documents
Culture Documents
Mechanical Engineering, University of Auckland, Private Bag 92019, Auckland, New Zealand
b
Silsoe Research Institute, Wrest park, Silsoe, Bedford MK45 4HS, UK
Received 10 December 2003; received in revised form 10 July 2004; accepted 14 July 2004
Available online 11 September 2004
Abstract
A quasi-steady method which uses observed mean pressure coefcients to predict the
expected peak positive or negative pressures is developed. It is shown that in the case of wall
pressures this involves calculating the joint probability of instantaneous wind direction and
gust dynamic pressure. With roof pressures the situation is more complex since the pressures
are also sensitive to elevation angles and so the joint probability also includes this angle.
Comparison of these predictions with observed data from the Silsoe 6 m cube show reasonable
agreement.
r 2004 Elsevier Ltd All rights reserved.
Keywords: Quasi-steady; Peak pressures; Cube; Wind loads; Full-scale
1. Introduction
Many wind loading standards, such as AS/NZS 1170.2:2002 [1], employ a quasisteady approach for the design of static structures. Cook [2] comments The quasisteady approach is a compromise which assumes that all the uctuations of load are
due to the gusts of the boundary layer; thus the contribution from the buildinggenerated turbulence is suppressed by this method. This leads to a design approach
Corresponding author. Tel. +64-9-3737999; fax: +64-9-3737479.
ARTICLE IN PRESS
1174
P.J. Richards, R.P. Hoxey / J. Wind Eng. Ind. Aerodyn. 92 (2004) 11731190
is the mean
where r is the air density, V^ is the expected peak wind speed and C p y
pressure coefcient, which is taken to be a function of mean wind direction alone.
Cook [2] further shows that the equivalent-steady-gust model is a severe
simplication of the quasi-steady vector model where the instantaneous pressure is
given by
1
pt rV 2 tC p y; b;
2
in terms of the mean wind speed U and the uctuating turbulence components u, v
and w.
The uctuating azimuth and elevation angles are given by
y0 y y arctanv=U u and b arctanw=U u:
ARTICLE IN PRESS
P.J. Richards, R.P. Hoxey / J. Wind Eng. Ind. Aerodyn. 92 (2004) 11731190
1175
of experimental data. In either case there are a number of approximations which are
commonly made. These include:
Ignoring the effects of the elevation angle: With this approximation the term Cp(y,b) is
reduced to Cp(y). This approximation is appropriate for pressures on the walls of
many buildings since these are relatively insensitive to the elevation angle. Data
collected on the Silsoe 6 m Cube showed that tilting the cube by up to 51 had very
little effect on the windward face vertical centreline pressures. Sharma [3] obtained
similar results in the University of Auckland wind tunnel by tilting a 1:50 scale model
of the Texas Tech Building. These experiments showed that the pressures in the centre
of the windward wall were insensitive to tilting although some points nearer the edge
did show some sensitivity, with values for |dCp/db| as high as 0.5 rad1. Even these
values are small in comparison with the derivatives obtained by the same method for
roof pressures. Sharma and Richards [4] quote roof pressure coefcient derivative as
high as 5.73 rad1 and show that retaining the relationship between pressure
coefcient and elevation angle can be important in the analysis of roof pressure
spectra. Letchford and Marwood [5] used a similar tilting model technique in the
University of Oxford Low Speed Wind Tunnel and obtained values of |dCp/db|
greater than 10 rad1 near the roof corner. They also found that incorporating the w
component turbulence term improved the comparison of quasi-steady predicted rms
pressures with measured values. It can therefore be concluded that where possible the
effects of elevation angle should be retained, particularly with roof pressures, however
it is recognised that retaining this term is often difcult since determining the
sensitivity of the pressure coefcient to elevation angles is often impossible.
Linearising the pressure coefficient-azimuth angle relationship: Many authors,
including Kawai [6], linearise the pressure coefcient-azimuth angle function.
While this is convenient it is very inaccurate near regions of maximum or
minimum pressure. In fact since the standard deviation of wind direction is often
of the order of 101, a linear range of 7301 is needed for this assumption to be
accurate, but this seldom occurs (see for example Fig. 3 later in this paper).
Richards et. al. [7] have suggested using a short Fourier series to represent this
function in order to avoid the need for linearisation.
Approximating the instantaneous pressure coefficient function by the mean pressure
coefficient: Richards et. al. [7] show that if an instantaneous function of the form
of Eq. (2) exists then the observed mean pressures have lower extreme values.
Hence if this function is approximated by measured mean pressure coefcients
then the expected maximum and minimum pressures will under-predict the likely
values. Methods such as those used by Richards et al. [7] or Banks and Meroney
[8], which seek to nd a function which when combined with direction variations
would lead to the observed mean values, can be used to give a more consistent
estimate of the instantaneous function.
ARTICLE IN PRESS
1176
P.J. Richards, R.P. Hoxey / J. Wind Eng. Ind. Aerodyn. 92 (2004) 11731190
where q(t) is the instantaneous dynamic pressure. The form of Cp(y), which is the
instantaneous variation of pressure coefcient with instantaneous direction, cannot
be easily determined since direct measurement requires q and y to be constant, but
doing this would remove the turbulent stresses that affect the ow eld. In practice
most standards assume that the instantaneous coefcient is equal to the mean
coefcient:
C p y C p y:
While this approach is simple and is a good rst order approximation, it is limited
and as pointed out in the previous section will tend to under-predict extreme values.
Eq. (7) is even less accurate when applied to the prediction of expected peak
pressures. This occurs because an extreme pressure may occur when an extreme
dynamic pressure q^ combines with a high pressure coefcient which occurs at an
Predicting the expected peak value is therefore a
angle near, but not necessarily at, y:
joint probability problem. If Eq. (6) is assumed to apply then for peak positive
pressures the probability of exceeding a particular threshold p+ is given by
y360
X
p
Qp4p
Q q4
PyDy;
8
C p y
y0
p
where Q q4 C py is the probability that the dynamic pressure will be strong enough
to produce the required pressure when combined with the instantaneous pressure
coefcient for that direction, P(y) is the probability density of wind directions which
will depend on the mean and standard deviation of wind directions during any
observation period and Dy is a narrow band of wind angles. Since the application of
Eq. (6) means that positive
pressures
can only occur with a positive Cp it is taken that
p
if Cp(y)o0 then Q q4 C py 0:
ARTICLE IN PRESS
P.J. Richards, R.P. Hoxey / J. Wind Eng. Ind. Aerodyn. 92 (2004) 11731190
1177
0.066h
Tap 17
Wind
Direction
Taps 22
and 23
Y
X
Taps 9
Taps 28
and 29
(a)
(b)
Fig. 1. (a) The Silsoe 6 m cube and (b) the position of the mid-height and roof pressure tappings.
ARTICLE IN PRESS
1178
P.J. Richards, R.P. Hoxey / J. Wind Eng. Ind. Aerodyn. 92 (2004) 11731190
collected for the ve pressure taps shown in Fig. 1(b) with a range of 12-min mean
wind directions and strengths. During some periods the cube was normal to the
prevailing wind while at other times the cube was rotated through 451. A total of 328
12-min blocks of data were recorded, however only those with mean dynamic
pressures greater than 20 Pa were used since the lower speed blocks tend to produce
unreliable results.
The data was processed in order to provide mean, standard deviation, maximum
and minimum values for the pressures, reference wind dynamic pressure and
direction. The pressure data has been reduced to coefcient form in the following
manner:
p
p^
p^
C p ; C^ p and C^ p :
q
q^
q^
10
6
X
bk sinky
a k cosky
11
k0
has been tted to the mean pressure coefcient data by using a least squares method.
The rms error was 0.045 in Cp.
Analysis of wind records at the site show that the direction variations are
approximately normally distributed about the mean angle such that during each 12min block
!
2
1
y y
Py p exp
:
12
2s2y
sy 2p
During the runs the standard deviation of wind directions sy ranged from
71(0.122 rad) to 181 (0.314 rad) with an average of 101(0.174 rad). Fig. 2 shows a
typical example of the distribution of wind directions about the mean.
The instantaneous pressure coefcient function has therefore been constructed by
following the method in Richards et al. [7], where
C p y
6
X
ak cosky bk sinky;
k0
with ak a k exp12 k2 s2y ; bk bk exp12 k2 s2y and sy =p/18 rad (101).
13
ARTICLE IN PRESS
P.J. Richards, R.P. Hoxey / J. Wind Eng. Ind. Aerodyn. 92 (2004) 11731190
1179
4
3.5
Measured data
Normal
3
2.5
2
1.5
1
0.5
0
-1
-0.8
-0.6
-0.4
-0.2
0.2
0.4
0.6
0.8
Fig. 2. Probability density function for the wind direction variations around the mean.
1.2
Tap 17
1
Tap 23 rotated
0.8
Tap 29 rotated
0.6
Tap 22 mirrored
0.4
Cp mean
Cp
0.2
Cp instant
0
-0.2
45
90
135
180
225
270
315
360
-0.4
-0.6
-0.8
-1
-1.2
Fig. 3 shows the variation of mean pressure coefcient with mean wind direction.
It may be observed that the combination of pressures from the ve tappings creates a
single consistent curve with relatively little scatter. The dashed line is the t to the
data (Eq. (11)) while the solid line is the corresponding estimated instantaneous
variation (Eq. (13)). The instantaneous curve reaches more extreme values in both
the positive and negative directions by about 0.1 in Cp.
In order to evaluate the probability of exceedence functions for the pressure in (8)
and (9) it is necessary to determine the probability of exceedence function for the
ARTICLE IN PRESS
P.J. Richards, R.P. Hoxey / J. Wind Eng. Ind. Aerodyn. 92 (2004) 11731190
1180
wind dynamic pressures. In other quasi-steady analyses, such as Banks and Meroney
[8], the wind speed has often been assumed to be normally distributed. Although the
overall wind speed statistics are reasonably matched by a normal distribution,
analysis of data recorded at Silsoe at a height of 6 m shows that the distribution of
wind speeds is slightly skewed towards increased wind speed. This means that if a
normal distribution model is used the probability of occurrence of the high winds
speeds, which are relevant for the prediction of extreme pressures, will be
underestimated. This is quite clear in Fig. 4, which shows a typical probability
density function for the wind speed on a log scale. A normal distribution can be seen
to match the central data but does not match either the low or high speed regions. As
a result a modied normal distribution was used.
This took the form
!
C V 1 b ab
C V a2
p exp
PC V
;
b1
2
2C 2b
CV
c 2p
Vc
14
10
1
0
0.2
0.4
0.6
0.8
1.2
1.4
1.6
0.1
0.01
0.001
Measured data
Normal
Modified Normal
0.0001
1.8
ARTICLE IN PRESS
P.J. Richards, R.P. Hoxey / J. Wind Eng. Ind. Aerodyn. 92 (2004) 11731190
1181
Fig. 5. Mapping of a normal distribution in y onto the semi-innite space for the velocity coefcient.
15
The corresponding probability density function for the dynamic wind pressure is
given by
s
PC V
2q
16
Pq q ; with C V
2
2
rV
2rV q
and the probability of exceedence obtained by numerically integration,
1
Qq4q
Zq
Pq dq:
17
Fig. 6 shows the resulting probability of exceedence values in terms of the dynamic
pressure coefcient Cq, which is the ratio of the particular dynamic pressure to the
mean dynamic pressure C q q=q:
With b E 0.23, the turbulence intensity c=0.19 and a probability of exceedence of
1 in 3000 (Q(4Cq)=0.00033), Eqs. (14), (16) and (17) show that the expected peak
to mean ratio for the dynamic pressure is 2.9, which is close to the average observed
ratio of 2.88.
ARTICLE IN PRESS
P.J. Richards, R.P. Hoxey / J. Wind Eng. Ind. Aerodyn. 92 (2004) 11731190
1182
10
1
0
0.5
1.5
2.5
3.5
Measured data
0.01
Normal
Modified Normal
Q(>Cq)=0.00033
0.001
Cq=2.9 2
0.0001
An Excel spreadsheet has been developed which uses the above equations to
predict the maximum and minimum pressure coefcients expected at Tap 17 with a
probability of exceedence of 1 in 3000 for each mean wind direction. These
calculations are shown as lines in Fig. 7 along with the observed mean, peak positive
(max) and peak negative (min) pressure coefcients. It should be noted that these
curves are derived from tting the mean pressure coefcient data and do not rely on
any measurements of peak values.
Fig. 7 shows that while the mean data points are clustered around the tted line,
the peak data show much greater scatter. Nevertheless it can be seen that in general
both the maximum and minimum data follow the trend suggested by the quasisteady predictions. It may be noted that if the mean pressure coefcient is negative
for a range of angles around that being considered (for example between 2201 and
3601) then the maximum positive pressure is near zero and the minimum pressure
coefcients are near to the mean values. Similarly if the mean pressure coefcient is
positive over a range around a particular direction (801 to 1001) then the minimum
pressure is near zero and the maximum pressure coefcient is close to the mean value
for that direction. These observations support the assumptions made earlier, in
association with the evaluation of Eq. (8) and (9), regarding the peak values when the
mean pressure coefcient is of opposite sign. Fig. 7 also shows that at times the
maximum and minimum values are signicantly different from the mean value. The
most extreme example of this occurs with a mean wind direction of 351, at which
point the mean pressure coefcient is near zero, however the maximum pressure
coefcient are in the range 0.50.9 and the minimum pressure coefcients are in the
range 0.75 to 1.5. It is reasonable to assume that the peak positive pressures
occur during strong gusts while the instantaneous wind direction has swung around
to about 551, where the instantaneous function gives a coefcient of 0.74, and that
the peak negative coefcients occur during strong gust and instantaneous wind
ARTICLE IN PRESS
P.J. Richards, R.P. Hoxey / J. Wind Eng. Ind. Aerodyn. 92 (2004) 11731190
1183
max
mean
min
Cp max
Cp mean
Cp min
Pressure Coefficient
0
0
60
120
180
240
300
360
-1
-2
-3
Fig. 7. A comparison of measured maximum, minimum and mean pressure coefcients (symbols)
with those predicted by a quasi-steady model (lines) for a point on the cube sidewall at mid height and
x/h=0.066.
directions around 151, where the instantaneous pressure coefcient is 0.9. The
extreme values hence depend on the joint probability of strong gusts and appropriate
wind directions.
The most noticeable difference between the expected and measured peak values
occurs with minimum pressures between 01 and 601 and around 1801. At these angles
Tap 17 is on the side of the cube and is affected by the separating and reattaching
ow. It is thought that these lower minimum peak pressures are the result of the
dynamic behaviour of this ow, which results in the periodic formation of intense
vortices, which are attached to the leading edge of the cube for a short time and then
shed into the general ow. This observation re-emphasis the point made at the end of
Section 1, that a quasi-steady model cannot be expected to account for every effect,
but if applied in a systematic manner can show what observations can be attributed
to quasi-steady processes and what should be attributed to other processes such as
building generated turbulence.
ARTICLE IN PRESS
P.J. Richards, R.P. Hoxey / J. Wind Eng. Ind. Aerodyn. 92 (2004) 11731190
1184
1.0
0.0
zero pitch
-0.2
Tap 9
0.0
mean Cp
mean Cp
0.5
-0.5
-1.0
-0.6
-0.8
zero pitch
2.5 deg pitch
5 deg pitch
-1.0
windward
roof
leeward
-1.2
-1.5
0
(a)
-0.4
12
60
18
75
(b)
position (m)
90
105
120
wind angle
Fig. 8. (a) The effect of tilting the Silsoe 6 m cube on vertical centreline plane mean pressure coefcients
for a wind direction parallel to the plane (901) and (b) the variation of mean pressure coefcients, over a
range of wind directions, for Tap 9 which is 3.5 m from the windward edge of the roof at 901.
noticeably, particularly at Tap 9. Fig. 8(b) shows the changes in mean pressure
coefcient at Tap 9 for a range of wind directions around 901. The results in this
gure suggest that for this point
@C p y; b
0:077C p y; 0 deg1
3:14C p y; 0 rad1
@b
Where the elevation effects are signicant then Eq. (9) becomes
y360
Xb
X b3s
p
Qpop
Q q4
PbDbPyDy;
C p y; b
y0 b3s
18
19
with the range of b set at three standard deviations either side of the mean, which is
assumed to be zero, in order to include all likely elevation angles. Data collected
at a height of 6 m suggested that the standard deviation of b is about 4.51 (0.078 rad).
Fig. 9 shows that the elevation angles are approximately normally distributed
around a mean of 01. Although it may be observed that the elevation angles may be
slightly skewed towards positive (upward angles) the use of a modied distribution in
this case is not warranted since the angles of primary interest are those around zero
rather than the extremes, as was the case with the dynamic pressure.
Evaluation of Eq. (19) is further complicated by the effects of the Reynolds shear
stress which means that the vertical velocity, and hence elevation angle, is partially
correlated with the wind speed, as illustrated in Fig. 10.
The data shown in Fig. 10 had the following statistics:
A linear t to the data in Fig. 10 has a gradient of 0.0704, which is also the ratio
of uw to the variance of the streamwise velocity su2. It may also be noted that if the
friction velocity, u, is calculated from the mean wind speed and a roughness length
ARTICLE IN PRESS
P.J. Richards, R.P. Hoxey / J. Wind Eng. Ind. Aerodyn. 92 (2004) 11731190
1185
Measured data
Normal
6
5
4
3
2
1
0
-0 .5
-0 .4
-0 .3
-0.2
-0 .1
0 .1
0 .2
0 .3
0 .4
0 .5
Fig. 9. Probability density function for the elevation angle in the free stream at a height of 6 m.
of 6 mm, then a simple logarithmic prole gives the friction velocity as 0.57 m/s and
the expected Reynolds stress as u2n 0:308 m2 =s2 ; which is close to the directly
measured value.
The instantaneous elevation angle is
w
w
b arctan
if b is small:
20
U u
U u
ARTICLE IN PRESS
1186
P.J. Richards, R.P. Hoxey / J. Wind Eng. Ind. Aerodyn. 92 (2004) 11731190
This means that the variation of the elevation angle which is correlated with the
streamwise uctuations, b, can be estimated from
u2
u
u2 V U
bn 2n
2n
:
21
V
su U u
su
Since the standard deviation of b is small and the range of elevation angles is always
centred on zero, then linearising the relationship between instantaneous pressure
coefcient and elevation angle is far more justied than it would be for the azimuth
angle y. While this linearisation is more justied, it is recognised that there may be
situations where it is not sufciently accurate. In more complex environments where
the variation of b is large, possibly due to upstream buildings, or the variation of Cp
with b is highly nonlinear, this method will be very approximate. Including these
approximations leads to the following expression for the quasi-steady minimum
pressure:
@C p y; b u2n V U
@C p y; b 0 r 2
b
V ;
p C p y; 0
22
@b
V
@b
2
s2u
where b0 is the random variation of elevation angle and is assumed to be normally
distributed.
The solution procedure involves solving Eq. (22), which is a quadratic in V, for a
given p, y and b0 combination, and then calculating the corresponding dynamic
pressure and hence obtaining the probability of exceedence for that situation. The
value of p with a probability of exceedence of 0.00033 is then extracted from the
calculations. The results of this procedure are shown in Fig. 11, where the mean
pressure data for Tap 9 is shown along with the mean pressure coefcient tted curve
and the corresponding instantaneous curve. Also shown are three estimates for the
expected variation of minimum pressure coefcient with direction. The three
estimates have been obtained by using only the rst term in Eq. (22), the rst two
terms and then all terms. Including only the rst term creates a situation where the
expected minimum curve is atter than the mean pressure coefcient curve. This
occurs because in regions where the mean pressure coefcient is less negative a more
negative pressure can be generated at an instant when a gust combines with a change
in wind direction that provides a more negative instantaneous coefcient. Inclusion
of the Reynolds stress term generally reduces the level of the expected minimum.
This occurs because a high wind speed is associated with a negative vertical velocity
and hence a negative elevation angle.
Fig. 8(a) shows that for Tap 9 the pressure coefcient became less negative as the
cube was tilted forward, this is equivalent to a negative elevation angle. Hence the
expectation is that high wind speeds will be correlated with lower coefcients and
hence the expected peak pressure is less negative. Including the nal term in Eq. (22)
results in the expected minimum becoming more negative. This may be slightly
surprising since the elevation angle is approximately normally distributed with a zero
mean. This means that the angle is just as likely to be positive as it is negative. Hence
with a linear variation the magnitude of the pressure coefcient is increased as often
ARTICLE IN PRESS
P.J. Richards, R.P. Hoxey / J. Wind Eng. Ind. Aerodyn. 92 (2004) 11731190
1187
Pressure Coefficient
-0.2
45
90
135
180
225
270
315
360
-0.4
-0.6
-0.8
-1
-1.2
-1.4
-1.6
-1.8
Mean Cp data
Mean Cp curve
Instataneous Cp
Cp min with no elevation terms
Cp min with Reynolds stress term
Cp min with full equation
-2
ARTICLE IN PRESS
P.J. Richards, R.P. Hoxey / J. Wind Eng. Ind. Aerodyn. 92 (2004) 11731190
1188
0
0
45
90
135
180
225
270
315
360
-0.2
-0.4
-0.6
-0.8
-1
-1.2
-1.4
Cp min data
Mean Cp
Q-S No elevation terms
Q-S Full equation
-1.6
-1.8
-2
Cp min/Cp mean
0
0
45
90
135
180
225
270
315
360
15% and possibly more, with some data in the range 25012901 showing a ratio as
high as 2. Hoxey et al. [9,10] have discussed the sources of the uncertainty in such
measured data and argue that the primary source is the measurement of the peak
ARTICLE IN PRESS
P.J. Richards, R.P. Hoxey / J. Wind Eng. Ind. Aerodyn. 92 (2004) 11731190
1189
-600
-400
-200
-200
-400
y = 0.9402x
R2 = 0.8746
-600
-800
Fig. 14. A comparison of the observed minimum pressure at roof Tap 9 on the Silsoe cube with the
expected values predicted by the full quasi-steady analysis.
dynamic pressure at a location some distance from the building. With the reference
anemometer about 25 m from the centre of the building then measurements [9] have
shown that it may be expected that the standard deviation of the differences between
measurements at the reference mast and at the centre of the building, but with the
building removed, would be about 10% of the typical dynamic pressure. Hence the
peak dynamic pressure measured at the reference mast may be much smaller than
that actually affecting the building. In wind tunnel testing this uncertainty may be
minimised by using extreme value analysis on multiple blocks of stationary data.
However, in full-scale testing one is relying on nature to provide the wind and so no
two 12 min blocks are truly similar, not even in regard to wind direction, and so
extreme value analysis would be questionable and is hence not attempted here.
In spite of these uncertainties, Fig. 14 shows that if the full quasi-steady pressure
coefcients are combined with the observed maximum wind dynamic pressures then
the expected minimum pressures at Tap 9 are well correlated with the measured
pressures although they are 6% too low.
5. Conclusions
A quasi-steady method which uses observed mean pressure coefcients to predict
the expected peak positive or peak negative pressures has been developed. It is shown
that in the case of wall pressures this involves calculating the joint probability of
instantaneous wind direction and gust dynamic pressure. With roof pressures the
situation is more complex since the pressures are also sensitive to elevation angles
and so the joint probability also includes this angle. Comparison of these predictions
with observed data from the Silsoe 6 m cube show reasonable agreement. Although
ARTICLE IN PRESS
1190
P.J. Richards, R.P. Hoxey / J. Wind Eng. Ind. Aerodyn. 92 (2004) 11731190
the data used in this study is derived from a single full-scale situation and hence may
not apply in general, it is believe by the authors that the principals outlined are
generally applicable and are not unique to the particular study.
References
[1] AS/NZS 1170.2:2002, Structural design actions, Part 2: Wind actions, Standards Australia, 2002.
[2] N.J. Cook, The designers guide to wind loading of building structures, Part 2: Static structures,
Butterworths, UK, 1990.
[3] R.N. Sharma, The inuence of internal pressure on wind loading under tropical cyclone conditions,
Ph.D. Thesis, University of Auckland, New Zealand, 1996.
[4] R.N. Sharma, P.J. Richards, The inuence of Reynolds stresses on roof pressure uctuations, J. Wind
Eng. Ind. Aerodyn. 83 (1999) 147157.
[5] C.W. Letchford, R. Marwood, On the inuence of v and w component turbulence on roof pressures
beneath conical vortices, J. Wind Eng. Ind. Aerodyn. 6971 (1997) 567577.
[6] H. Kawai, Pressure uctuations on square prisms-application of strip and quasi-steady theories,
J. Wind Eng. Ind. Aerodyn. 13 (1983) 197208.
[7] P.J. Richards, R.P. Hoxey, B.S. Wanigaratne, The effect of directional variations on the observed
mean and rms pressure coefcients, J. Wind Eng. Ind. Aerodyn. 54/55 (1995) 359367.
[8] D. Banks, R.N. Meroney, The applicability of quasi-steady theory to pressure statistics beneath rooftop vortices, J. Wind Eng. Ind. Aerodyn. 89 (2001) 569598.
[9] R.P. Hoxey, P.J. Richards, G.M. Richardson, A.P. Robertson, J.L. Short, The folly of using extremevalue methods in full-scale experiments, J. Wind Eng. Ind. Aerodyn. 60 (1996) 109122.
[10] R.P. Hoxey, A.P. Robertson, P.J. Richards, How have full-scale measurements improved the
reliability of wind-loading codes? Dick Marshalls contribution to full-scale measurements of wind
effects. 11th International Conference on Wind Engineering, Lubbock, Texas, USA, July 2003,
pp. 29-48.